Pydantic AI Documentation Skill

Overview

This skill provides guidance for using Pydantic AI - a Python agent framework for building production-grade Generative AI applications. Pydantic AI emphasizes type safety, dependency injection, and structured outputs.

Key Concepts

Agents

Agents are the primary interface for interacting with LLMs. They contain:

• Instructions: System prompts for the LLM
• Tools: Functions the LLM can call
• Output Type: Structured datatype the LLM must return
• Dependencies: Data/services injected into tools and prompts

Models

Pydantic AI supports multiple LLM providers via model identifiers.

All models that supports tool-calling can be used with pydantic-ai-skills.

Tools

Two types of tools:

• @agent.tool: Receives RunContext with dependencies
• @agent.tool_plain: Plain function without context

Toolsets

Collections of tools that can be registered with agents:

• FunctionToolset: Group multiple tools
• MCPServerTool: Model Context Protocol servers
• Third-party toolsets (ACI.dev, etc.)

Instructions

1. Fetch Full Documentation

For comprehensive information, fetch the complete Pydantic AI documentation: https://ai.pydantic.dev/llms.txt

This contains complete documentation including agents, tools, dependencies, models, and API reference.

2. Quick Reference

Basic Agent Creation

from pydantic_ai import Agent

agent = Agent('openai:gpt-5.2')
result = agent.run_sync('What is the capital of France?')
print(result.output)

Agent with Tools

from pydantic_ai import Agent, RunContext

agent = Agent('openai:gpt-5.2', deps_type=str)

@agent.tool
def get_user_name(ctx: RunContext[str]) -> str:
    """Get the current user's name."""
    return ctx.deps

result = agent.run_sync('What is my name?', deps='Alice')

Structured Output

from pydantic import BaseModel
from pydantic_ai import Agent

class CityInfo(BaseModel):
    name: str
    country: str
    population: int

agent = Agent('openai:gpt-5.2', output_type=CityInfo)
result = agent.run_sync('Tell me about Paris')
print(result.output)  # CityInfo(name='Paris', country='France', population=...)

Dependencies

from dataclasses import dataclass
from pydantic_ai import Agent, RunContext

@dataclass
class MyDeps:
    api_key: str
    user_id: int

agent = Agent('openai:gpt-5.2', deps_type=MyDeps)

@agent.tool
async def fetch_data(ctx: RunContext[MyDeps]) -> str:
    # Access dependencies via ctx.deps
    return f"User {ctx.deps.user_id}"

Using Toolsets

from pydantic_ai import Agent
from pydantic_ai.toolsets import FunctionToolset

toolset = FunctionToolset()

@toolset.tool
def search(query: str) -> str:
    """Search for information."""
    return f"Results for: {query}"

agent = Agent('openai:gpt-5.2', toolsets=[toolset])

Async Execution

import asyncio
from pydantic_ai import Agent

agent = Agent('openai:gpt-5.2')

async def main():
    result = await agent.run('Hello!')
    print(result.output)

asyncio.run(main())

Streaming

from pydantic_ai import Agent

agent = Agent('openai:gpt-5.2')

async with agent.run_stream('Tell me a story') as response:
    async for text in response.stream():
        print(text, end='', flush=True)

3. Common Patterns

Dynamic Instructions

@agent.instructions
async def add_context(ctx: RunContext[MyDeps]) -> str:
    return f"Current user ID: {ctx.deps.user_id}"

System Prompts

@agent.system_prompt
def add_system_info() -> str:
    return "You are a helpful assistant."

Tool with Retries

@agent.tool(retries=3)
def unreliable_api(query: str) -> str:
    """Call an unreliable API."""
    ...

Testing with Override

from pydantic_ai.models.test import TestModel

with agent.override(model=TestModel()):
    result = agent.run_sync('Test prompt')

4. Installation

# Full installation
pip install pydantic-ai

# Slim installation (specific model)
pip install "pydantic-ai-slim[openai]"

5. Best Practices

1. Type Safety: Always define deps_type and output_type for better IDE support
2. Dependency Injection: Use deps for database connections, API clients, etc.
3. Structured Outputs: Use Pydantic models for validated, typed responses
4. Error Handling: Use retries parameter for unreliable tools
5. Testing: Use TestModel or override() for unit tests